Anthropic revealed a new capability on Monday, enabling Claude users to access interactive applications directly within the chatbot environment.
Reflecting Anthropic’s business-oriented approach, the initial set of applications largely comprises workplace instruments like Slack, Canva, Figma, Box, and Clay, with Salesforce integration anticipated shortly. Each application will establish a logged-in service instance that Claude can access, empowering users to perform actions such as sending Slack messages, creating charts, or retrieving cloud files, contingent on the enabled apps.
In a blog post introducing this feature, Anthropic stated, “Analyzing data, designing content, and managing projects are all enhanced by a dedicated visual interface.” They added, “When combined with Claude’s intelligence, you can work and iterate more rapidly than either could achieve independently.”
This new functionality is offered to Pro, Max, Team, and Enterprise subscribers, excluding free users. Qualified users can enable these tools via claude.ai/directory.
This system bears resemblance to OpenAI’s Apps system, which debuted in October and similarly facilitates interactive third-party tools. Both app integration frameworks are founded upon the Model Context Protocol (MCP), an open standard that Anthropic unveiled in 2024. MCP commenced app support in November, a collaborative effort between both firms.
These new applications will gain significant potency when integrated with Claude Cowork, a versatile agent tool Anthropic introduced last week. Based on Claude Code, Cowork enables users to delegate complex, multi-stage tasks involving extensive, open-ended datasets — operations that previously demanded terminal commands. When combined with the new app functionality, Cowork could access cloud-based files or ongoing work projects. For instance, Cowork might update a marketing graphic in Figma or utilize fresh data from the company’s Box account.
While apps are not initially available within Cowork, Anthropic indicated that this integration is “coming soon.”
Agentic systems inherently carry a degree of unpredictability, and Anthropic’s safety guidelines for Cowork advise users to diligently oversee the agent and refrain from bestowing any superfluous permissions.
The company advises, “Exercise caution when granting access to sensitive information such as financial documents, credentials, or personal records.” They further suggest, “Instead of providing broad access, contemplate establishing a dedicated working folder for Claude.”